1# RECORD 1908d2ee0b0995f38129f51d7446a3626531b8d62# TITLE: Mind the Anticipation Gap: A Difference-in-Differences Analysis of the Elizabeth Line’s Impact on London Housing Prices3# YEAR: 2026.0 VENUE: Vietnamese Economic Review DOI: 10.67913/vner/2026.1.1.14# AUTHORS: T. Nguyen5# PAGES: 15 CANDIDATE TABLE PAGES (1-based): [6, 8, 10, 11]6# ABSTRACT: This paper estimates the causal impact of London’s Elizabeth Line on residential property prices using a Difference-in-Differences framework applied to 1,602,487 transactions from the UK Land Registry over 2000–2024. Treated properties within five distance bands of stations are compared to a baseline beyond 10 km, separately for the construction window (2007–2015) and the post-opening window (2022–2024). During construction, prices within 1 km of stations rose by 4.9% relative to control, with substantially larger gains of 7.7–8.1% in the 2–10 km bands, indicating broad anticipatory capitalisation and spillover effects. Following the May 2022 opening, station-adjacent prices fell by 2.0%, consistent with overshooting during the construction phase, while outer-band prices remained unchanged. Effects are concentrated in zones previously under-served by Transport for London. The findings have implications for transport appraisal, value-capture timing, and equity-oriented infrastructure planning.789===== PAGE 1 (score 9, numeric-paren 0, table-words 0, var-words 9) =====1011Vietnamese Economic Review12https://doi.org/10.67913/vner/2026.1.1.113Mind the Anticipation Gap:14A Difference-in-Differences Analysis15of the Elizabeth Line’s Impact16on London Housing Prices17Tien Phong Nguyen 1181 King’s Business School, King’s College London19Abstract20This paper estimates the causal impact of London’s Elizabeth Line on residential21property prices using a Difference-in-Differences framework applied to 1,602,48722transactions from the UK Land Registry over 2000–2024.23Treated properties24within five distance bands of stations are compared to a baseline beyond 10 km,25separately for the construction window (2007–2015) and the post-opening win-26dow (2022–2024). During construction, prices within 1 km of stations rose by274.9% relative to control, with substantially larger gains of 7.7–8.1% in the 2–2810 km bands, indicating broad anticipatory capitalisation and spillover effects.29Following the May 2022 opening, station-adjacent prices fell by 2.0%, consis-30tent with overshooting during the construction phase, while outer-band prices31remained unchanged. Effects are concentrated in zones previously under-served32by Transport for London. The findings have implications for transport appraisal,33value-capture timing, and equity-oriented infrastructure planning.34Keywords: Elizabeth Line; Transport infrastructure; House prices.35JEL Classification: R41, R31, R42, H5436137Introduction38Urban transport infrastructure shapes residential property values primarily39through accessibility and commuting-time effects (Gibbons & Machin 2005).40Empirical work consistently documents premiums for properties near transit,41with effect sizes mediated by network design, anticipation behaviour, and pre-42existing accessibility (Mohammad et al. 2017; Ahlfeldt 2013; Billings 2011). The43Crossrail project—rebranded the Elizabeth Line on opening—offers an unusually44well-defined natural experiment: an £18 billion east–west rail link spanning 11845km from Reading to Shenfield, formally approved in 2007, under construction46Corresponding author(s): Tien Phong Nguyen, King’s Business School, King’s College London,47tien_phong.nguyen@kcl.ac.uk481495051===== PAGE 6 (score 19, numeric-paren 0, table-words 6, var-words 4) =====5253654Mind the Anticipation Gap55is the single strongest cross-sectional predictor of log price (R2 = 0.89), substan-56tially exceeding the explanatory power of macroeconomic indicators. Inflation,57interest rates, and GDP growth show statistically significant but quantitatively58small bivariate relationships with prices, consistent with the dominance of lo-59calised factors in urban housing markets (Ahlfeldt 2013).60Table 1: Descriptive statistics by distance band, 2000–202461Distance band62Observations63Share64(%)65Mean66log67price68SD log69price70Mean71distance72(km)73School74(km)75TfL76station77(km)78Within 1 km79127,028807.938112.71820.88830.64841.48852.54861–2 km87194,3818812.138912.67900.80911.51922.43933.87942–5 km95480,8379630.019712.76980.81993.381001.711012.971025–10 km103433,84910427.0710512.771060.731077.221080.431091.80110More than 10 km111366,39211222.8611312.641140.6611513.511160.371172.59118Total1191,602,487120100.0012112.721220.761236.291241.131252.64126Notes: Sample includes residential transactions in Greater London and outer boroughs contain-127ing Elizabeth Line stations, 2000–2024. Distance refers to Euclidean distance to the nearest128Elizabeth Line station; School and TfL station refer to nearest non-Elizabeth amenity. Log129prices are natural logarithms of nominal sale prices in pounds sterling.1304131Empirical Strategy1324.1133Difference-in-Differences Specification134The baseline DiD model is given by equation 1:135ln(Price)it = α + β(Treatedi × Postt) + γXit + δt + ηi + εit136(1)137where Treatedi is a binary indicator for properties within 2 km of an Elizabeth138Line station, Postt is an indicator for the post-treatment period (either post-2007139construction or post-2022 opening), Xit is a vector of property and macroeco-140nomic controls, δt are year fixed effects absorbing common shocks, and ηi are141property fixed effects absorbing unobserved time-invariant heterogeneity. The142coefficient of interest β captures the causal impact of proximity to the Elizabeth143Line, expressed in approximate percentage terms.144The richer specification replaces the binary treatment with distance-band indi-145cators:146ln(Price)it = α +1474148∑149d=1150βd(DistanceBanddi × Postt) + γXit + δt + ηi + εit151(2)152where DistanceBanddi takes values for the within 1 km, 1–2 km, 2–5 km, and 5–15310 km bands relative to the omitted reference of more than 10 km. The vector βd154155156===== PAGE 7 (score 16, numeric-paren 1, table-words 3, var-words 9) =====157158Tien Phong Nguyen1597160traces the spatial profile of the treatment effect. Equations 1 and 2 are estimated161separately for two windows. The construction window covers 2000–2015 with162Post = 1 from 2007. The opening window covers 2016–2024 with Post = 1 from1632022. This split isolates anticipatory capitalisation, when the line was authorised164but not running, from post-opening adjustment, when accessibility gains were165physically realised.1664.2167Parallel Trends168Identification requires that, absent treatment, treated and control properties169would have followed parallel price paths.170The assumption is assessed with171placebo regressions estimating year-by-year coefficients in the pre-treatment172windows (2000–2006 for the construction phase; 2016–2021 for the opening173phase). Pre-construction estimates show closely aligned price trajectories across174distance bands.175For example, in 2001 the log price coefficient for properties176within 1 km of future stations is 0.022, virtually identical to the 1–2 km band177(0.025), and the alignment holds through 2006.178In the opening pre-period179(2016–2021), modest divergences emerge from 2020 onward—properties within1801 km record a small negative coefficient of -0.030 in 2021 (p = 0.032), reflecting181anticipatory adjustment as the operational launch approached.182This pattern183is consistent with prior literature documenting anticipatory effects ahead of184major transport openings (Gibbons & Machin 2005; Agostini & Palmucci 2008)185and reflects rational expectations rather than a violation of parallel trends.186Robustness checks below confirm that the main DiD estimates are not driven by187these end-of-window deviations.1884.3189Event Study190To trace the temporal profile of treatment effects, I estimate the event study in191equation 3:192ln(Price)it = α +193−2194∑195T=−q196βT(Treatedi × DT197t ) +198m199∑200T=0201βT(Treatedi × DT202t )203+ γXit + δt + ηi + εit204(3)205where DT206t indicates calendar year T, and the year preceding treatment (T =207−1) is omitted as the baseline. The βT coefficients capture year-specific devi-208ations between treated and control properties. The event-study design enables209tests for pre-trends, identifies the timing of capitalisation, and characterises post-210treatment dynamics.2114.4212Robustness213Several robustness exercises are conducted. First, property-type indicators (de-214tached, semi-detached, terraced, flat), estate type, and amenity distances are215216217===== PAGE 8 (score 27, numeric-paren 11, table-words 5, var-words 4) =====2182198220Mind the Anticipation Gap221added to the baseline specification. Second, borough socioeconomic controls are222added for the 2017–2024 sub-window. Third, heterogeneity is tested by inter-223acting the treatment with an indicator for pre-existing TfL access, defined as a224non-Elizabeth TfL station within 1 km. Fourth, for the post-2017 sub-window225I compare fixed-effects (FE) and random-effects (RE) specifications. The Haus-226man test is inconclusive because of a non-positive-definite variance-covariance227matrix—a common outcome when within-property variation is limited (Moham-228mad et al. 2017). The Breusch–Pagan LM test (p < 0.001) and a moderate rho229of 0.36 motivate FE as the primary specification while reporting RE estimates230as a secondary check, consistent with prior DiD work (Gibbons & Machin 2005;231Billings 2011). Standard errors are clustered at the property level throughout.2325233Results2345.1235Construction Phase, 2007–2015236Table 2: DiD estimates: construction phase, 2007–2015237Binary (1)238Distance bands (2)239Treated × Construction240−0.003241(0.002)242Within 1 km × Construction2430.049244(0.003)2451–2 km × Construction2460.053247(0.002)2482–5 km × Construction2490.081250(0.002)2515–10 km × Construction2520.077253(0.002)254Inflation rate2550.320256−0.006257(0.017)258(0.001)259Interest rate260−6.0552610.060262(0.243)263(0.005)264GDP growth rate2650.7102660.032267(0.025)268(0.003)269Property and year fixed effects270Yes271Yes272Observations273541,796274541,796275Notes: Dependent variable is the natural logarithm of nominal sale276price. Standard errors clustered at the property level shown in paren-277theses. Reference category in column (2) is properties more than 10278km from an Elizabeth Line station. Coefficients on “Yes”-valued rows279indicate inclusion of the corresponding fixed effects; standard errors280omitted for these rows.281Table 2 reports DiD estimates for the construction window. The binary specifi-282cation in column (1) yields a small and only marginally significant coefficient of283−0.003 (p = 0.063), suggesting little aggregate price movement when spatial284285286===== PAGE 9 (score 11, numeric-paren 0, table-words 3, var-words 5) =====287288Tien Phong Nguyen2899290heterogeneity is collapsed. Column (2) replaces the binary treatment with the291four distance-band interactions and tells a substantively different story. All four292coefficients are positive and significant at the 1% level. Properties within 1 km293of an Elizabeth Line station appreciated by approximately 4.9% relative to the294more-than-10-km reference group; the 1–2 km band gained 5.3%. The largest295effects, however, are in the mid-range bands: 8.1% in the 2–5 km band and 7.7%296in the 5–10 km band. Macroeconomic controls enter with the expected signs.297The non-monotonic distance gradient is economically interpretable. Mid-range298zones may have benefited from a combination of expected accessibility gains and299limited exposure to construction disamenities such as traffic, noise, and ground300works concentrated near station sites. Properties in the immediate vicinity were301more exposed to these short-term costs, partially offsetting the anticipated long-302term benefits. The pattern echoes findings from Agostini & Palmucci (2008) and303Dubé et al. (2013), who document larger anticipation effects in zones marginally304removed from new infrastructure. The construction-phase event study traces this305pattern over time: year-specific coefficients for properties within 2 km of stations306turn positive shortly after the 2007 announcement, rise through the early 2010s,307and reach their peak in 2014–2016, foreshadowing the correction documented308below. Figure 2 presents the event-study coefficients for the full 2000–2024 hori-309zon.310Figure 2: Event-study estimates for properties within 2 km of an Elizabeth Line station, 2000–3112024. The y-axis is the year-specific log-price coefficient relative to the reference year (2006).3125.2313Post-Opening Phase, 2022–2024314Table 3 reports DiD estimates for the opening window. The binary specification315in column (1) gives a coefficient of −0.014 (p = 0.001), indicating that prices316within 2 km of a station are 1.4% lower than control prices following the May3172022 launch. Column (2) disaggregates by distance band: properties within 1 km318319320===== PAGE 10 (score 26, numeric-paren 11, table-words 3, var-words 7) =====32132210323Mind the Anticipation Gap324decline by 2.0% relative to the reference (p = 0.005); the 1–2 km band declines325by 1.2% (p = 0.024). Coefficients for the 2–5 km and 5–10 km bands are statis-326tically indistinguishable from zero. The post-opening profile is the spatial mirror327of the construction-phase result: the correction is sharpest in the bands that expe-328rienced the largest pre-opening anticipation, particularly the inner bands where329construction-disamenity offsets had limited the apparent uplift.330Table 3: DiD estimates: post-opening phase, 2022–2024331Binary (1)332Distance bands (2)333Treated × Opening334−0.014335(0.004)336Within 1 km × Opening337−0.020338(0.007)3391–2 km × Opening340−0.012341(0.005)3422–5 km × Opening343−0.001344(0.004)3455–10 km × Opening346−0.003347(0.004)348Inflation rate349−0.002350−0.002351(0.001)352(0.001)353Interest rate3540.0193550.019356(0.002)357(0.002)358GDP growth rate3590.0143600.014361(0.001)362(0.001)363Property and year fixed effects364Yes365Yes366Observations367170,093368170,093369Notes: Dependent variable is the natural logarithm of nominal sale370price. Standard errors clustered at the property level shown in paren-371theses. Reference category in column (2) is properties more than 10372km from an Elizabeth Line station. The Opening indicator equals one373from 2022 onwards.374The opening-phase event study confirms this interpretation.375The treated–376control gap, having peaked in 2014–2016, narrows steadily from 2017 onward377and turns modestly negative for inner-band properties from 2022. By 2023–2024,378treated coefficients are close to zero, consistent with the market settling at or379slightly below the level implied by realised accessibility gains. The pattern is380consistent with rational expectations and information efficiency:381anticipated382benefits were priced in well before service commenced, leaving little scope383for further gains and some scope for correction of mid-2010s overshooting.384Comparable post-opening reversion has been documented around Dubai Metro385openings (Mohammad et al. 2017) and in commuter rail studies (Dubé et al.3862013).387388389===== PAGE 11 (score 19, numeric-paren 6, table-words 4, var-words 4) =====390391Tien Phong Nguyen392113935.3394Heterogeneity by Pre-existing Transport Access395A central question for value-capture and equity policy is whether the Elizabeth396Line’s price effects vary with pre-existing accessibility. Equation 4 tests this by397adding interactions between the treatment and an indicator TfLi for whether a398property is within 1 km of a non-Elizabeth TfL station:399ln(Price)it = α + β1(Treatedi × Postt) + β2(TfLi × Postt)400+ β3(Treatedi × TfLi × Postt) + γXit + δt + ηi + εit401(4)402Table 4 reports the construction-phase and opening-phase estimates from equa-403tion 4. For properties near new Elizabeth Line stations but without prior TfL cov-404erage, the construction-period effect is negative (−0.038, p < 0.001), indicating405that early construction disruption was costly for these previously under-served406locations. The triple interaction with prior TfL access is positive and significant407(0.021, p < 0.001), indicating that pre-existing TfL connectivity partly insulated408nearby properties from construction-period disamenities. In the opening window,409the triple interaction is positive but not statistically significant (0.015, p = 0.25),410indicating that by 2022 the marginal value of additional TfL access at Elizabeth411Line stations had essentially dissipated.412Table 4: Heterogeneity by pre-existing TfL access413Construction (1)414Opening (2)415Treated × Post416−0.038417−0.021418(0.004)419(0.010)420TfL × Post4210.105422−0.034423(0.001)424(0.003)425Treated × TfL × Post4260.0214270.015428(0.006)429(0.013)430Property and year fixed effects431Yes432Yes433Macro and amenity controls434Yes435Yes436Observations437541,796438170,093439Notes: TfL is an indicator equal to one for properties within 1 km440of a non-Elizabeth TfL station. Treated indicates properties within4412 km of an Elizabeth Line station. Standard errors clustered at the442property level shown in parentheses. Column (1) covers 2007–2015;443column (2) covers 2022–2024.444Event-study estimates by sub-group reinforce this. Properties within 1 km of445an Elizabeth Line station that were already well served by TfL show steady appre-446ciation through 2004–2010 followed by gradual decline; the Elizabeth Line adds447little marginal value, consistent with a saturation hypothesis. Properties within 1448km that were not previously connected to TfL show delayed but stronger growth,449peaking around 2014 through 2018, before stabilising post-opening. The het-450erogeneity is consistent with diminishing marginal returns to accessibility and451with international evidence that infrastructure adds the most value in previously452under-served areas (Ahlfeldt & Wendland 2011; Koster et al. 2014).453454455===== PAGE 12 (score 14, numeric-paren 0, table-words 1, var-words 12) =====45645712458Mind the Anticipation Gap4595.4460Macroeconomic and Property-Level Controls461Macroeconomic controls perform consistently with theory but should be inter-462preted cautiously because year fixed effects absorb most national-level variation.463Inflation and interest rates are negatively associated with prices in both windows;464GDP growth is positively associated. Magnitudes are larger in the construction465window, reflecting wider macroeconomic variation across 2000–2015 (including466the 2008–2009 recession). In the post-2017 window, broader shocks—the Brexit467referendum, the COVID-19 pandemic, and the 2022 monetary tightening—appear468to have suppressed further appreciation in zones where anticipatory capitalisa-469tion had already occurred, consistent with Vergara-Perucich (2024) and Hilber &470Vermeulen (2016).471Property-type controls cannot be identified in the FE specification because prop-472erty type is time-invariant. In the RE specification estimated for the 2017–2024473sub-window for which all controls are observed, flats show the largest negative474coefficient relative to the omitted base (−0.75, p = 0.003), consistent with flats475in already-dense central locations being most exposed to overshooting and sub-476sequent correction. Detached homes show −0.63 (p = 0.003); terraced homes477show −0.40 (p = 0.002). These results should be read as conditional compar-478isons rather than causal effects, given the omitted within-property variation in479FE.4806481Discussion482The estimates support three substantive conclusions.483First, the Elizabeth484Line generated economically meaningful capitalisation effects, but these were485concentrated in the construction phase rather than the operational phase.486Forward-looking buyers priced expected accessibility gains long before service487commenced, consistent with the rational-expectations capitalisation framework488(Agostini & Palmucci 2008; Ahlfeldt 2013).489Second, the spatial profile of490anticipation is non-monotonic:491mid-range zones (2–10 km) experienced the492largest construction-period gains, plausibly because they combined expected493accessibility benefits with limited exposure to construction disamenities. Third,494the post-opening correction is concentrated in inner bands, suggesting that an-495ticipation in those zones overshot the realised improvement in net accessibility496once disamenities, congestion, and competing routes are taken into account. The497overshooting–correction pattern echoes findings around Dubai Metro openings498(Mohammad et al. 2017) and other rail studies (Dubé et al. 2013).499The heterogeneity by pre-existing TfL access has direct implications for500transport-led equity and value-capture policy.501The Elizabeth Line generated502larger and more durable price responses in zones not previously served by TfL,503consistent with the diminishing marginal returns to accessibility predicted by504bid-rent theory and confirmed in international evidence (Ahlfeldt & Wendland5052011; Koster et al. 2014). This argues for prioritising future transit investment506in under-served catchments where capitalisation effects—and the social welfare507gains they signal—are largest. Three caveats are important. First, distance-based508